VoiceBridge: General Speech Restoration with One-step Latent Bridge Models

VoiceBridge is a new AI model that can restore high-quality 48kHz speech from various types of audio distortions using a single one-step process. The model uses a latent bridge approach with an energy-preserving variational autoencoder and transformer architecture to handle multiple speech restoration tasks simultaneously.

• →VoiceBridge introduces a unified one-step latent bridge model capable of handling diverse speech restoration tasks with a single architecture.
• →The model can efficiently reconstruct 48kHz fullband speech from various distortions without requiring distillation.
• →An energy-preserving variational autoencoder enhances waveform-latent space alignment across different energy levels.
• →The system demonstrates superior performance across both in-domain tasks like denoising and out-of-domain tasks like refining synthesized speech.
• →Joint training of the latent bridge model, decoder, and discriminator transforms the system from a denoiser into a generator.